In this repository a neural network was written in Java which classifies MNIST data.
What does the data consist of?
"The MNIST database (Modified National Institute of Standards and Technology database) is a large database of handwritten digits that is commonly used for> training various image processing systems. [..] The MNIST database contains 60,000 training images and 10,000 testing images. "
The following Java code was programmed as a proof-of-concept prototype without much emphasis on documentation (recognizable by the fact that comments and methods were written in German). It served the understanding of the Gradient Descent Algorithm.
Windows:
javac NeuronalesNetz.java
java NeuronalesNetz
Linux:
make
This network has an accuracy of ~ 85%.
Which is not very good, convolutional Neural Networks have an accucary of ~99%, but since this was about learning gradient descent algorithm, it's fine
In order to make this code work MNSIT Dataset is necessary. I've uploaded the files here:
https://www.file-upload.net/download-14510874/mnist_test.txt.html
https://www.file-upload.net/download-14510879/mnist_train.txt.html
just move them into the source code folder.